Loading documentation…
Loading documentation…
Design clients that remain within application request limits.
Treat rate-limit response headers and documented 429 errors as authoritative. Coordinate concurrency across workers instead of letting each process retry independently.
Cache repeated reads and avoid polling faster than your product freshness requirement.
An application key has a shared request limit. A per-process counter cannot see calls from other workers, scheduled jobs, or deployment replicas. Use a shared limiter or deliberately partition work so aggregate traffic remains bounded.
When the API rejects a request for rate limiting, stop immediate retries. Use the documented reset metadata to schedule later work and add randomized jitter so a fleet does not resume at once.
Track operation, outcome, and wait duration without using raw keys as labels. Alert on sustained rejections and queue age; a single bounded rejection can be normal backpressure.
Caching and request coalescing often reduce rate pressure more safely than increasing concurrency. See Caching and Prepare for production.